Home » Software Dev » 524 n8n nodes in one free dataset to stop agent hallucinations

524 n8n nodes in one free dataset to stop agent hallucinations

Cagri Sarigoz

Published May 18, 2026

Building AI agents that write n8n workflows has one specific failure mode: the agent picks a node that looks plausible, invents an operation that does not exist, and produces a workflow that imports cleanly and then breaks at runtime. Hard to catch. Expensive to debug.

Developer Artyom Rabzonov published a structured catalog of every n8n node to HuggingFace to solve exactly that problem. It covers 524 nodes, every operation, every credential type, and a top-level properties schema. Free. CC-BY-4.0.

️ What Each Row Contains

Every row in the dataset maps to one node and includes:

• node_name: internal ID (e.g. slack, lmChatOpenAi)
• display_name: the label you see in the UI
• categories and subcategories: taxonomy values
• operations_supported: the actual operation values, not inferred ones
• credentials_required: exact credential type names
• properties_schema: JSON describing top-level property descriptors
• source_package: either nodes-base or @n8n/nodes-langchain
• github_permalink: pinned link to the .node.ts source file

Format is JSON and Parquet (Snappy). The dataset updates monthly.

The Agent Pipeline This Enables

The intended usage pattern is RAG over a tool catalog. Embed every row using description, operations, and credentials. At plan time, retrieve the top N nodes relevant to the user request. Hand the agent only those rows. Validate the emitted workflow JSON against the properties schema before deploy.

The developer includes a code example that filters nodes by operation using the HuggingFace datasets library:

from datasets import load_dataset

ds = load_dataset("automatelab/n8n-nodes-catalog")["train"]

messaging = ds.filter(
    lambda r: "message" in (r["operations_supported"] or [])
)
for row in messaging:
    print(row["node_name"], row["credentials_required"])

Numbers Worth Knowing

The catalog split landed at 431 nodes from nodes-base and 93 from @n8n/nodes-langchain. The langchain package is not a small footnote. A non-trivial number of nodes have an empty operations_supported list: those are root nodes like LLMs, vector stores, and output parsers where the operation abstraction does not apply. Useful to know if your planner filters by operation.

⚠️ Caveats

• The properties schema is a top-level summary. For deep parameter shapes, use the github_permalink.
• Multi-version nodes report only the default version. The source link covers full version history.
• The CC-BY-4.0 license covers the catalog additions. The n8n source itself is governed by n8n’s own license.

A browsable index is also available at automatelab.tech if you want to explore nodes without loading the full dataset.

Cagri Sarigoz( Founder )

I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.

With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.

At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.

And more than all else, I’m a father to a (always) little girl and a husband.

Contact me at [email protected] for assistance.